W3C home > Mailing lists > Public > public-swbp-wg@w3.org > February 2005

RE: [WNET],[OEP] OntoWordNet. A new large OWL ontology

From: Christopher Welty <welty@us.ibm.com>
Date: Fri, 25 Feb 2005 08:29:25 -0500
To: "Govoni, Darren" <DGovoni@mcdonaldbradley.com>
Cc: "Aldo Gangemi" <a.gangemi@istc.cnr.it>, "Uschold, Michael F" <michael.f.uschold@boeing.com>, public-swbp-wg@w3.org, public-swbp-wg-request@w3.org
Message-ID: <OF4BF2475E.79ACD308-ON85256FB3.00493E6E-85256FB3.004A1927@us.ibm.com>
Cyc has a pretty good lexical ontology.  Almost all the cyc terms have a 
corresponding set of lexical entries.  Of course the camel-cased names 
themselves cannot be used for any kind of lexical information - they are 
not intended to be anything other than unique names that convey some 
meaning to human readers - but they are linked to an MT that does contain 
(usually quite good and reliable) lexical information.  There is also a 
mapping in Cyc to WordNet.

I really have no idea why you're mentioning Cyc in this context.  The 
discussion should stay focused on Wordnet in OWL. 

-Chris

Dr. Christopher A. Welty, Knowledge Structures Group
IBM Watson Research Center, 19 Skyline Dr., Hawthorne, NY  10532     USA   
 
Voice: +1 914.784.7055,  IBM T/L: 863.7055, Fax: +1 914.784.7455
Email: welty@watson.ibm.com, Web: 
http://www.research.ibm.com/people/w/welty/



"Govoni, Darren" <DGovoni@mcdonaldbradley.com> 
Sent by: public-swbp-wg-request@w3.org
02/25/2005 07:37 AM

To
"Aldo Gangemi" <a.gangemi@istc.cnr.it>, "Uschold, Michael F" 
<michael.f.uschold@boeing.com>
cc
<public-swbp-wg@w3.org>
Subject
RE: [WNET],[OEP] OntoWordNet. A new large OWL ontology






I would add that a 'good' lexical ontology adheres to an _internally 
consistent_ formal language with formal parsing rules that work across the 
ontology. In the case of wordnet, its the english language (ok, I know 
about all the grammatical nuances of natural language that could be 
inconsistent). Look at something like CYC. Sure, it has concepts that are 
_mostly_ human readable, but it by NO means follows a consistent format. 
Example. I have an easy concept 'Terrorism'. Its parsable and conforms to 
natural language. It's a simple word. You also will find 'TerrorismEvent' 
or wierd things like 'DefiningMT', which are not words and do not follow a 
strict lexicon or set of parse rules (using camel-case parsing might get 
you a long way in CYC, but since its not internally consistent it breaks 
at some point). Therefore, I would NOT refer to CYC as a lexical ontology, 
even though most of the concepts can be interpreted by a human - its not 
without a bit of human processing to decipher them and those rules may 
vary.

Darren


-----Original Message-----
From: Aldo Gangemi [mailto:a.gangemi@istc.cnr.it]
Sent: Fri 2/25/2005 3:59 AM
To: Uschold, Michael F; Govoni, Darren
Cc: public-swbp-wg@w3.org
Subject: Spam:RE: [WNET],[OEP] OntoWordNet. A new large OWL ontology

At 12:52 -0800 24-02-2005, Uschold, Michael F wrote:
>Aldo:
>
>Thanks for the definition of a lexical ontology.
>It seems to be rooted not in the nature of the beast, but more in
>the nature of how the beast is created, and to an extent, what it is
>intended to be used for.
>IF the starting point is a bunch of words or phrases that one wants
>to have a model of, *and*
>IF if the intended use entails [somethign like] language parsing, or
>lexical analysis of some sort... then one is more likley to call it
>a lexical ontology.
>
>On retrofitting 'lexical' terms to an otherwise non-lexical ontology...
>I suppose one shoudl always be able to come up with phrases that
>correspond to the meaning of the concepts in any ontology.
>Im not sure it helps or explains much to say this makes it a
>'lexical ontology', since the phrases are kind of invented, not
>arising from an existing lexicon.
>

yes, that's a point: how much ontology organization depends on
linguistic normalization? maybe three layers:

1) whatever lexicalization the designer likes in order to catch its
intuition ("kind of invented", actually: "subjective")
2) lexicalization emerging from learning procedures (not invented,
but neither "normalized" or "conventional", actually: "observable")
3) conventional or normalized lexicalization through lexicographic
procedures (actually: "agreed")

I think each layer has its own original motivation, as well as
ontologies have. I've a taste for 2) whenever possible, but when
strong negotiation is required, 1) is better, and there is so much
existing in 3)!
Let alone the encrusted meta-talk from philosophy, linguistics,
lexicography, etc., which has its own independent life.

A


>Mike
>
>
>-----Original Message-----
>From: Aldo Gangemi [mailto:a.gangemi@istc.cnr.it]
>Sent: Thursday, February 24, 2005 8:14 AM
>To: Uschold, Michael F; Govoni, Darren
>Cc: public-swbp-wg@w3.org
>Subject: RE: [WNET],[OEP] OntoWordNet. A new large OWL ontology
>
>I agree with Mike: reliability and the right analytic detail for the
>task at hand are the essential requisites for a good ontology.
>How to get them, it's another story. How to measure them, still
>another. WordNet can be good for its generality, which is also its
>weakness. That's why the TF contains an activity aimed at indicating
>how to use WordNet to create something else. In the meantime,
>WordNet seems to be useful in many cases.
>
>Concerning lexical ontologies, there is a quite straightforward
>definition: if the elements of an ontology (classes, properties, and
>individuals, possibly axioms) depend primarily on the acceptance of
>existing lexical entries, the ontology can be called "lexical".
>WordNet, formal or not, it's such a case.
>
>But one can force this statement, by saying that if one is able to
>build a comprehensible paraphrase in some natural language of each
>ontology element, then that's a linguistically-sound ontology. Which
>holds for most (if not all) ontologies.
>
>Therefore, "lexical" depends on the agreement of lexicographers. In
>fact, if we use an ontology learning technique from corpora, and
>state the boundaries of lexical units according to dynamic
>functional properties, such an ontology would be very different from
>a "lexical" ontology.
>
>Cheers
>Aldo
>
>At 7:49 -0800 24-02-2005, Uschold, Michael F wrote:
>>I have not seen any good definitions clarifying the difference
>>between a 'lexical ontology' vs. other kinds of ontologies.
>
>"ontology=taxonomy with relations" is as good or better than any
>other view of an ontology, for the sake of discussion.
>
>However, the more important issue is not what is or is not an
>ontology, but rather, what purpose any 'ontology-like artifact'
>serves.
>Insofar as WN hyper/hyponymy links are inaccuarte, WN will not be
>reliable for supporting tasks that require reliable taxonomic
>inference.
>Insofar as WN lacks relation, WN will not provide good support for
>tasks that require them.
>
>Mike
>
>-----Original Message-----
>[MFU] 
>From: Govoni, Darren [mailto:DGovoni@mcdonaldbradley.com]
>Sent: Wednesday, February 23, 2005 10:52 AM
>To: Uschold, Michael F; Aldo Gangemi; public-swbp-wg@w3.org
>Cc: brian.mcbride@hp.com; welty@us.ibm.com; schreiber@cs.vu.nl;
>glottolo@ilc.cnr.it; jjc@hplb.hpl.hp.com; swick@w3.org;
>danbri@w3.org; guarino@loa-cnr.it; oltramari@loa-cnr.it;
>ciaramita@loa-cnr.it
>Subject: RE: [WNET],[OEP] OntoWordNet. A new large OWL ontology
>
>Hi,
>    I haven't chimed in much recently, but I've been working with
>WordNet, CYC and various ontologies here at McDonald Bradley for a
>while. I even made an OWL version of WordNet about a year ago.
>
>    To the point on whether Wordnet is an ontology, I offer my
>opinion based on this, rather simple definition of ontology
>(forgetting where I first learned it). ontology=taxonomy with
>relations.
>
>I see WordNet as something of a lexical ontology. I lacks some of
>the machine esoteric, existential abstractions that something like
>CYC has. Mileage varies on the utility of that, IMO.
>
>Insomuch as the various OWL models we use manifest in much the same
>form (nodes or concepts connected by relations), our WordNet OWL
>model is every bit identical in nature to our CYC one. In our
>graphical ontology browser, they have exactly the same structure.
>That is, a graph (and RDF triples). Hard core ontologists will claim
>an ontology is a more formalized class/property/abstraction model
>(like CYC) whereas WordNet dismisses generic abstractions in favor
>of lexical symbols (i.e. human readble). Personally, I don't find
>the difference to be terribly salient. Plato basically posited words
>to be abstract symbols anyway.
>
>What we've found is that regardless of what you call it most
>ontologies are suitable up to a point before extending, modifying or
>mapping them to accomplish a goal is necessary. But that is not
>really a measure of 'ontology-ness', IMO.
>
>Just my thoughts.
>
>Darren
>
>Senior Architect
>McDonald Bradley
>
>
>
>-----Original Message-----
>From: public-swbp-wg-request@w3.org on behalf of Uschold, Michael F
>Sent: Wed 2/23/2005 1:04 PM
>To: Aldo Gangemi; public-swbp-wg@w3.org
>Cc: brian.mcbride@hp.com; welty@us.ibm.com; schreiber@cs.vu.nl;
>glottolo@ilc.cnr.it; jjc@hplb.hpl.hp.com; swick@w3.org;
>danbri@w3.org; guarino@loa-cnr.it; oltramari@loa-cnr.it;
>ciaramita@loa-cnr.it
>Subject: RE: [WNET],[OEP] OntoWordNet. A new large OWL ontology
>
>Here are a few thoughts about WordNet and ontologies gathered during
>last week's Dagstuhl Workshop on: Machine Learning for the Semantic Web
>
>The use of WN is more and more prevalent these days, especially among
>those working with ontologies.
>However, WN is designed as a lexical resource, not an ontology; it was
>never intended to be an ontology.
>
>Anyone who tries to use WN as an ontology quickly discovers that many of
>the hyper/hyponymy links are not proper taxonomic links at all.  This
>raises the question as to whether and when WN should be used as an
>ontology at all.
>
>If you try to use a knife as a can-opener - beware. It sort of works
>kinda, but you need to be careful.
>
>I dont have an opinion on this, but thought I'd report on these views
>that I learned of.
>
>It would be useful to have something to say on this point in the TF
>outputs.
>
>Mike
>
>
>
>         -----Original Message-----
>         From: Aldo Gangemi
>[<mailto:a.gangemi@istc.cnr.it>mailto:a.gangemi@istc.cnr.it]
>         Sent: Wednesday, February 16, 2005 12:35 AM
>         To: public-swbp-wg@w3.org
>         Cc: Uschold, Michael F; brian.mcbride@hp.com; welty@us.ibm.com;
>schreiber@cs.vu.nl; glottolo@ilc.cnr.it; jjc@hplb.hpl.hp.com;
>swick@w3.org; danbri@w3.org; guarino@loa-cnr.it; oltramari@loa-cnr.it;
>ciaramita@loa-cnr.it
>         Subject: [WNET],[OEP] OntoWordNet. A new large OWL ontology
> 
> 
>         Hi all,
>
>         second message for new [WNET] files.
>
>         This message is about a new version of the WordNet datamodel
>that we started modelling months ago. First versions were encoded by
>Guus Schreiber and Brian McBride. This version (3) has been enlarged,
>commented, and checked after the original WordNet specifications by me.
>It's downloadable from:
><http://www.w3.org/2001/sw/BestPractices/WNET/wordnet_datamodel.owl>
http://www.w3.org/2001/sw/BestPractices/WNET/wordnet_datamodel.owl.
>Extensive documentation from original sources, and about the work
>carried out, is contained in the OWL file.
>
>         Best
>         Aldo
>
>
>         --
>
>
>
>         Aldo Gangemi
>         Research Scientist
>         Laboratory for Applied Ontology
>         Institute for Cognitive Sciences and Technology
>         National Research Council (ISTC-CNR)
>         Via Nomentana 56, 00161, Roma, Italy
>         Tel: +390644161535
>         Fax: +390644161513
>         a.gangemi@istc.cnr.it
> 
>         *******************
>         !!! please don't use the old gangemi@ip.rm.cnr.it
>         address, because it is under spam attack
>
>
>--
>
>
>
>Aldo Gangemi
>Research Scientist
>Laboratory for Applied Ontology
>Institute for Cognitive Sciences and Technology
>National Research Council (ISTC-CNR)
>Via Nomentana 56, 00161, Roma, Italy
>Tel: +390644161535
>Fax: +390644161513
>a.gangemi@istc.cnr.it
>
>*******************
>!!! please don't use the old gangemi@ip.rm.cnr.it
>address, because it is under spam attack


--



Aldo Gangemi
Research Scientist
Laboratory for Applied Ontology
Institute for Cognitive Sciences and Technology
National Research Council (ISTC-CNR)
Via Nomentana 56, 00161, Roma, Italy
Tel: +390644161535
Fax: +390644161513
a.gangemi@istc.cnr.it

*******************
!!! please don't use the old gangemi@ip.rm.cnr.it
address, because it is under spam attack
Received on Friday, 25 February 2005 13:30:01 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 8 January 2008 14:17:15 GMT